A driver in New York pulls up a Grubhub recruitment ad on a phone screen. Six words: "Earn up to $40 per hour." The company's own data showed one in 1,000 drivers ever hit that rate. That number was not an estimate. It was a recruitment ad priced at the 99.9th percentile.

Most people who have run a small operation have seen this phrase somewhere. A franchise pitch. A platform partner program. An affiliate deal. "Earn up to" sits at the top of the page, and the brain does the rest.

There is a name for this in the literature. Howard, Hardisty, Griffin, and Wang published a study in early 2025 called "Income Prediction Bias in the Gig Economy." They ran four studies across two gig platforms. What they found: workers do not overpredict how much they earn per hour. They overpredict how many hours they will work.

The rate on the screen felt right. A driver saw $40 and thought, "I could hit that on a good night." The error was not in the rate. It was in the hours. A worker planned to drive thirty hours a week. They drove twelve. The math broke before the first shift started.

The problem is not that workers lacked grit. The problem is that "earn up to" replaced real earnings data with a ceiling number, and the worker's own bias filled in the gap.

Grubhub lost 50 percent of its driver base every seven to eight months. Half the workforce, gone, every seven to eight months.

That churn created a loop. Drivers left because the pay did not match the ad. The company needed more drivers. More drivers meant louder ads. Louder ads meant higher ceiling numbers. Higher ceiling numbers brought in more workers who would also leave.

This is not a motivation problem. It is a measurement problem. The whole cycle ran on a gap the company built and kept open on purpose.

The better principle: when someone shows you an earnings ceiling, find the median. Once that is clear, three moves follow from it.

Not the top. Not the average. The median. If a franchise, a platform, or a partner program leads with "top earners make X," ask what the middle earner makes. If they cannot give you that number, or will not, that tells you more than the ad ever did. This one question costs nothing and changes the full shape of the decision.

The Howard study found that the bias lives in the hours, not the rate. Before you sign on to anything that pays by the hour or by the unit, write down how many hours you plan to work in month three. Then cut that number by 40 percent. That figure is closer to what the data show people actually do.

High turnover among participants is the tell. If half the people in a program leave within a year, the earnings claim was a recruiting number, not an operating number. Ask how long the typical participant stays. If the answer is vague, the number is bad.

Running this for a few weeks does something the original pitch never did:

You start to see which programs price their ads at the ceiling and which show you the median up front. You notice which franchise documents bury the 50th-percentile line on page forty. You catch the moment your own hour forecast starts to drift upward. And you begin to treat "earn up to" as a label for how they recruit, not a promise of what you will earn.

At the end of the next quarter, ask three things.

→ Which opportunity showed me the median before I asked for it?
→ Which pitch led with the ceiling and went quiet when I asked for the middle?
→ Where did my own hour forecast turn out to be wrong, and by how much?

That is the difference between a pitch that sounds right and a system that proves itself.

The FTC settled with Grubhub for a $140 million judgment. The company could not pay it. The number dropped to $25 million. Of that, $23.8 million went out as checks to 640,038 drivers and diners. The median payout was $51.

A driver who saw "Earn up to $40 per hour" on a phone screen got a check for fifty-one dollars. The next time any pitch leads with "up to," you now know what that word costs the person who believes it.